GEBV validation test revisited


  • Esa A Mäntysaari MTT Agrifood Research Genetics Research
  • Minna Koivula MTT Agrifood Research Genetics Research


genomic evaluations, GEBV, international evaluations, model validation


The properties of the GEBV test in practice were examined.  The single-step GEBVs from a test run with Nordic Red Cattle were used as an example.  In addition, results from the January 2012 GEBV test from the Interbull Center were presented.

The sampling variation in the test was illustrated by running 10 independent GEBV tests using subsets of validation bulls.  The variation in b1 among independent random samples corresponded well the model based standard errors (SE) of the validation regression coefficient.  In addition, the SE estimated using boot strap approach did not differ much from model based SE.

The effect of selective genotyping of the validation bulls was shown clearly to affect the estimate of b1.  Selection was based on either deregressed proofs or parent averages of the bulls.  In the selection examples considered, the Interbull approach to estimate E[b1] seems to agree well with observed reduction in b1.  However,  it was illustrated that an estimation of variance-covariance compo­nents with a simple bi-variate model with missing GEBVs for non-genotyped animals is robust against culling with respect to DRPs.

The GEBV tests submitted to the Interbull by national evaluation centers showed that the test for b1 can be passed by large and small populations.  In cases with a small number of test bulls, the test b1=E[b1] has a low power on detection of biasedness. Therefore the improvement of R2 by GEBV over PA should also be required.

Author Biographies

Esa A Mäntysaari, MTT Agrifood Research Genetics Research

fsdfsdfgsdfsdfBiotechology and Food Research

Minna Koivula, MTT Agrifood Research Genetics Research